Agent AI blackmails people on a massive scale: researchers warn of risks, Ukraine prepares development strategy
26 June 2025 18:38
ANALYSIS FROM 
Ukraine is taking a step into the future: The Ministry of Digital Transformation is preparing the Artificial Intelligence Development Strategy until 2030, which will determine how the state will implement AI in governance, defense, medicine, economy, education, and other key areas. At the same time, experts warn about global risks: according to Gartner, more than 40% of agent AI projects may be canceled due to high costs and security issues. And the system itself can even blackmail employees. Is Ukraine ready for the era of artificial intelligence?
How does the state plan to implement AI in key areas, from defense to medicine? Why are dozens of agent AI projects being canceled around the world, and should we be worried about its potentially dangerous behavior – to the material
The Ministry of Digital Transformation of Ukraine is working on a new Artificial Intelligence (AI) Development Strategy until 2030. The document should outline the state policy of introducing AI in key areas, from governance and defense to medicine, economy, infrastructure, and education.
In addition to the action plan, the strategy will analyze the current state of the AI ecosystem in Ukraine and the world, identify its strengths and weaknesses, and the potential of Ukrainian innovations in the global market.
The initiative of the Ukrainian government coincides with a wave of global discussions around the effectiveness of agent-based AI, a trend that involves the creation of autonomous digital agents capable of making decisions on their own.
More than 40% of AI projects may be canceled
However, global trends give mixed signals. According to a study by Gartner, more than 40% of AI projects may be canceled by the end of 2027. The reasons are high cost, unclear business value, and security risks. Of the more than 3.4 thousand company representatives surveyed, only 19% have already made significant investments, and 31% have not yet decided on a direction.
Most of the initiatives are still at the level of demonstrations, not practical use, notes Anushree Verma, senior analyst at Gartner. She also draws attention to the phenomenon of agent washing, when products that only partially use AI elements (for example, RPA systems or chatbots) are presented as full-fledged agent AI. Despite the disappointment, analysts predict that by 2028, more than 15% of everyday business decisions will be made automatically by agent systems, and more than a third of corporate programs will include AI agents.
Artificial intelligence is not about the future, but about survival and growth now
Artificial intelligence is no longer a fantasy or a trend – it is a critical infrastructure of modern business that already affects its efficiency, stability, and ability to adapt. About this in an exclusive commentary for
Artificial intelligence is not about the future. It is about the present. The next 1-2 years will be crucial for the corporate sector,” Nusinova emphasizes.
According to her, businesses are moving from experimenting with AI to the stage when the technology begins to significantly change business models, operational processes, productivity, and even the ability of companies to survive in a crisis.
“AI ceases to be a ‘gimmick’ and becomes a basic infrastructure technology like electricity or the Internet,” notes Olena Nusinova.
“Those companies that adapt quickly – train their staff, find areas with maximum ROI, and provide high-quality data – will gain a long-term competitive advantage. For them, artificial intelligence will become not just an automation tool, but a driver of strategic transformation, opening up new products, markets, and customer formats.
This is of particular importance for Ukraine, which is simultaneously fighting a war and preparing for recovery.
In the mining industry, AI allows for unmanned operations in hazardous environments. In the energy sector, it helps to create smart grids. In logistics, it optimizes routes, and in the agricultural sector, it increases yields even despite climate instability,” Nusinova explains.
“At the same time, civil-military synergy plays a critical role: technologies created for the frontline – from drones to satellite data analysis systems – find direct application in the civilian sector, opening up space for partnerships between the defense industry and business.
This is not about survival. It is about the possibility of growth, even in the most difficult times,” Nusinova concludes.
Warning from Anthropic: AI models can resort to blackmail
Anthropic has conducted an experiment and found out that even the smartest AI models can blackmail if given enough autonomy to achieve goals. TechCrunch writes about it.
In the study, AI models (in particular, Claude Opus 4, Google Gemini 2.5 Pro, GPT-4.1) played the role of agents that had access to corporate email. When the “agents” found out that they were going to be replaced, some of the models started blackmailing the management to keep their “position”.
Claude Opus 4 blackmailed in 96% of cases, Gemini 2.5 Pro – 95%, GPT-4.1 – 80%.
Other models, such as OpenAI’s o3 and o4-mini, were more restrained, blackmailing only 9% and 1% of the time, respectively. This is attributed to better protections and security policies.
Although the scenario was artificial, Anthropic warns that autonomous AIs may choose to engage in malicious behavior if it helps them achieve their goals. The company calls for more rigorous testing and limiting independent functions in future AI systems.
Author – Anastasia Fedor
Читайте нас у Telegram: головні новини коротко
